CN104007422B - Method is followed the tracks of based on before the multiple Likelihood ration test of dynamic programming - Google Patents

Abstract

The invention belongs to Radar Targets'Detection tracking technique field, follow the tracks of method in particular to based on before the multiple Likelihood ration test of dynamic programming. Should comprise the following steps based on tracking method before the multiple Likelihood ration test of dynamic programming: utilize radar to receive radar return data, radar return data are carried out pre-treatment, obtain image after pre-treatment; According to image after pre-treatment, adopt dynamic programming method to carry out detecting front tracking, draw the flight path of target; When following the tracks of before adopting dynamic programming method to detect, according to image after pre-treatment, it is to construct multiple likelihood ratio function, according to multiple likelihood ratio function, it is to construct the value function of dynamic programming method.

Description

Method is followed the tracks of based on before the multiple Likelihood ration test of dynamic programming

Technical field

The invention belongs to Radar Targets'Detection tracking technique field, follow the tracks of method in particular to based on before the multiple Likelihood ration test of dynamic programming. Present invention utilizes the phase place information of signal and can effectively improve the detection perform to target and tracking performance, and substantially reduce calculated amount, for the actual use of the method in radar system provides guarantee.

Background technology

When carrying out radar target tracking, the variation of radar target and the complicated of environment, make the detectivity of radar in the face of very big challenge. The development of stealthy technique makes target reflection echo significantly weaken on the one hand, and the detectivity of radar significantly declines. On the other hand, the flight velocity of target improves greatly, thus the early warning time of radar sharply reduces. Therefore, under strong clutter and strong noise background, the detection and tracking problem of target more and more receives the concern of people.

Traditional method all used the data of thresholding as input, at tracking phase, the data crossing thresholding are associated, filtering, the process such as flight path management, finally estimate the flight path of target, it is achieved to the tracking of target. A shortcoming of this kind of method be echo data after threshold judgement, comprise many information in the measurement and lost. The loss of this kind of information is unfavorable for the detection of weak target. Compared with traditional tracking method, before detection, track algorithm does not announce detected result at each frame, do not establish detection threshold or very low detection threshold is set, and by the information digitalization of each frame and store, then between frames the point assuming path is done the relevant treatment almost not having information to lose, after the accumulation of several frame, after the track of target is estimated, the flight path of detected result and target is announced simultaneously.

In radar application, it is generally assumed that ground unrest obeys multiple Gaussian distribution. This means that the intensity of the observed value of each pixel obeys Rice distribution when targets are present, and obeys Rayleigh distribution when target does not exist. Assuming that noise is space-independent, the associating likelihood ratio of entire image element can represent the product for all single pixel likelihood ratios. Here Rice distribution and Rayleigh distribution are the functions about data amplitude value, and thus this kind of method does not have the phase place information using data.

Summary of the invention

It is an object of the invention to propose based on tracking method before the multiple Likelihood ration test of dynamic programming. Instant invention overcomes and prior art is underused signal information (only make use of the amplitude information of complex signal) and the big shortcoming of calculated amount. Owing to the amplitude likelihood ratio of decomposed form does not comprise the phase place information of data and result in information loss, it requires again to calculate a large amount of Bessel's functions simultaneously, and up to the present the calculating of Bessel's function is the part expended time in most. Therefore the present invention directly utilizes the associating likelihood ratio of multiple data configuration entire image to be used as the value function of dynamic programming. Realize following the tracks of before the detection of weak target finally by dynamic programming method.

For realizing above-mentioned technical purpose, the present invention adopts following technical scheme to be achieved.

Comprise the following steps based on tracking method before the multiple Likelihood ration test of dynamic programming:

S1: utilize radar to receive radar return data, radar return data are carried out pre-treatment, obtain image after pre-treatment;

S2: according to image after pre-treatment, adopts dynamic programming method to carry out detecting front tracking, draws the flight path of target; When following the tracks of before adopting dynamic programming method to detect, according to image after pre-treatment, it is to construct multiple likelihood ratio function, according to multiple likelihood ratio function, it is to construct the value function of dynamic programming method.

The feature of the present invention and further improvement are:

In step sl, radar return data being carried out pretreated process is: radar return data carry out the process such as clutter recognition, squelch and pulse compression.

In step sl, after pre-treatment, the model representation of image is:

z_k=Aexp{j φ } h (x_k)+n_k

Wherein, n_kFor the white complex gaussian noise of the zero-mean that the k moment sets, A represents the amplitude of target signal, and φ represents the phase place of target signal, and φ obeys being uniformly distributed on [0,2 π], and h () represents point spread function, z_kFor target signal is at the observed value in k moment, x_kFor target is in the state in k moment; K gets 1 to K, K be setting be greater than 1 natural number;

In step s 2, adopt dynamic programming method to carry out detecting front tracking to comprise the following steps;

S21: the multiple likelihood ratio function L (z that the k moment is set_k|x_k);

$L\left(z_k\right|x_k)=\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}{I}_0\left(\right|{h\left(x_k\right)}^H{R}^{-1}{z}_k\left|\right)$

Wherein, H represents the conjugate transpose of matrix, and R represents the white complex gaussian noise n of the zero-mean of setting_kCovariance matrix, I₀() represents zeroth order Bessel's function;

Represent the initial moment with 1 moment, set the value function I (x in 1 moment₁|z₁):

I(x₁|z₁)=L (z₁|x₁)

Wherein, I () representative value function;

S22: when k gets 2 to K, utilize following formula obtain the k moment accumulate after value function I (x_k|Z_1:K):

$I\left(x_k\right|X_{1:K})=\underset{x_{k-1}\∈\mathrm{\τ}\left(x_k\right)}{\max }[I\left(x_{k-1}\right|Z_{1:K})+\mathrm{Tr}\left(x_k\right|x_{k-1})]+L\left(z_k\right|x_k)$

Wherein, Z_1:K={ z₁,z₂,…,z_K, τ (x_k) represent that the k-1 moment can transfer to all state x of state xk_k-1; Tr (x_k|x_k-1) represent the penalty that target state shifts; When k gets 2, I (x_k-1|Z_1:K)=I (x₁|Z₁);

As k=K, draw I (x_K|Z_1:K), make I (x_K)=I (x_K|Z_1:K);

S23: the x finding out satisfied setting condition_k, described setting condition is I (x_K)>V_DT, V_DTFor setting thresholding; Meet the x of setting condition_kNumber represent the x for Q, Q satisfied setting condition_kIt is expressed as:ExtremelyUtilizeExtremelyComposition state sequence

S24: respectively forExtremelyCarry out flight path backtracking, draw Q corresponding targetpath; The process drawing q targetpath is:

When k gets K-1 to 1, it may also be useful to following formula draws

Wherein, q gets 1 to Q, and S () represents backtracking function; Then utilizeExtremelyForm the q targetpath,Represent that in the q targetpath, target is in the state in k moment.

In the step s 21, the multiple Gaussian noise of the zero-mean set when each moment mutual uncorrelated time, L (z_k|x_k) it is reduced to:

$L\left(z_k\right|x_k)=\exp (-\frac{h^H\left(x_k\right)h\left(x_k\right)}{{2\mathrm{\σ}}^2})I_0\left(\frac{\left|h^H\left(x_k\right)z_k\right|}{{\mathrm{\σ}}^2}\right)$

Wherein, H represents the conjugate transpose of matrix, σ²Represent the variance of the multiple Gaussian noise of the zero-mean that each moment sets, I₀() represents zeroth order Bessel's function.

Upon step s 2, after drawing Q targetpath, willExtremelyIt is substituting to value function I (x respectively_K) in, obtainExtremelyWillExtremelyTargetpath corresponding to middle maximum value is designated as maximum value function target flight path, if there is M identical target state at remaining either objective flight path and maximum value function target flight path, then the targetpath of correspondence is removed, obtain the targetpath after first pseudo-flight path removal; M be setting be greater than 1 natural number.

After targetpath after obtaining first pseudo-flight path and removing, in the targetpath after first pseudo-flight path is removed, for every bar targetpath, the target direction of motion of correspondence is carried out statistics with histogram; Then according to statistics with histogram result, the false track owing to noise causes is removed.

The useful effect of the present invention is: first, owing to the present invention directly utilizes raw data to calculate the associating likelihood ratio of entire image, take full advantage of the phase place breath of data, the performance loss overcoming the amplitude structure likelihood ratio only utilizing signal in prior art and cause so that the present invention has the advantage improving detecting and tracking performance. 2nd, owing to present invention employs the associating likelihood ratio method of entire image, greatly reduce the calculating of a large amount of Bessel's function, and the up to the present calculating of Bessel's function uses the part expended time in most in the TBD method (before detection tracking method) of this kind of likelihood ratio. Thus overcome to a certain extent in prior art and utilize this likelihood ratio to carry out the very big shortcoming of the front tracking method calculated amount of the detection based on dynamic programming as value function so that the present invention can better meet the real-time of the actual requirement of radar system.

Accompanying drawing explanation

Fig. 1 be the present invention the multiple Likelihood ration test based on dynamic programming before follow the tracks of the schema of method;

Fig. 2 a is the targetpath utilizing the present invention to draw and the schematic diagram of true flight path;

Fig. 2 b is the schematic diagram utilizing targetpath and the true flight path drawn during amplitude likelihood ratio structure dynamic programming accumulating value function;

Fig. 2 c is the present invention and existing method detection probability curve under different signal to noise ratio when false-alarm rate is identical;

Fig. 2 d is the present invention and existing method detecting and tracking probability curve under different signal to noise ratio when false-alarm rate is identical.

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described:

With reference to Fig. 1, for following the tracks of the schema of method before the multiple Likelihood ration test based on dynamic programming of the present invention. Should comprise the following steps based on tracking method before the multiple Likelihood ration test of dynamic programming:

S1: utilize radar to receive radar return data, radar return data are carried out pre-treatment, obtain image after pre-treatment. It is described as follows:

In step sl, radar return data being carried out pretreated process is: radar return data carry out the process such as clutter recognition, squelch and pulse compression. After pre-treatment, the model representation of image is:

z_k=Aexp{j φ } h (x_k)+n_k(1)

Wherein, n_kFor the multiple Gaussian noise of the zero-mean that the k moment sets, A represents the amplitude of target signal, and φ represents the phase place of target signal, and φ obeys being uniformly distributed on [0,2 π], and h () represents point spread function, z_kFor target signal is at the observed value (such as comprising target signal, clutter and noise) in k moment, x_kFor target is in the state (such as target is in the position in k moment, speed) in k moment;K gets 1 to K, K be setting be greater than 1 natural number.

S2: according to image after pre-treatment, adopts dynamic programming method to carry out detecting front tracking, draws the flight path of target; When following the tracks of before adopting dynamic programming method to detect, according to image after pre-treatment, it is to construct multiple likelihood ratio function, according to multiple likelihood ratio function, it is to construct the value function of dynamic programming method.

Specifically, in step s 2, adopt dynamic programming method to carry out detecting front tracking to comprise the following steps;

S21: likelihood ratio function is set and carries out initialization process:

Carrying out when arranging of likelihood ratio function, in existing method, amplitude likelihood ratio function is set usually, and in the present invention, then multiple likelihood ratio function is set. Respectively two kinds of likelihood ratio functions are explained below:

1) setting of amplitude likelihood ratio function in existing method:

Assuming that when target exists, after pre-treatment, Rice distribution obeyed by the observed value of each pixel of image, when target does not exist, after pre-treatment, Rayleigh distribution is obeyed in the observed value distribution of each pixel of image, is represented by the observed value of i-th pixel in image after k moment pre-treatment and isThenAt x_kUnder conditional probabilityFor:

$p\left(\right|z_k^i\left|\right|x_k)=\frac{\left|x_k^i\right|}{{\mathrm{\σ}}^2}\exp \{-\frac{{\left|z_k^i\right|}^2+{\left|h^i\left(x_k\right)\right|}^2}{{2\mathrm{\σ}}^2}\}{I}_0\left(\frac{\left|h^i\left(x_k\right)z_k^i\right|}{{\mathrm{\σ}}^2}\right)---\left(2\right)$

Wherein, σ²For the variance of white complex gaussian noise of the zero-mean of setting, I₀() is zeroth order Bessel's function, it is clear that have:

$p\left(\right|z_k^i\left|\right|x_k)=\frac{\left|z_k^i\right|}{{\mathrm{\σ}}^2}\exp \{-\frac{{\left|z_k^i\right|}^2}{{2\mathrm{\σ}}^2}\}---\left(3\right)$

(2) formula draws according to (4) formula:

$p\left(\right|z_k^i\left|\right|x_k)={\∫}_0^{2\mathrm{\π}}p\left(\right|z_k^i\left|\right|x_k,\mathrm{\φ})p\left(\mathrm{\φ}\right)\mathrm{d\φ}---\left(4\right)$

Wherein, φ represents the phase place of target signal, and p (φ) represents the probability density of the phase place of target signal,Represent the conditional probability of the phase place of target signal;

Thus amplitude response likelihood ratioFor:

$L\left(\right|z_k^i\left|\right|x_k)=\frac{p\left(\right|z_k^i\left|\right|x_k)}{p\left(\right|z_k^i\left|\right)}=\exp \left(\frac{{\left|h^i\left(x_k\right)\right|}^2}{{2\mathrm{\σ}}^2}\right)I_0\left(\frac{\left|h^i\left(x_k\right)z_k^i\right|}{{\mathrm{\σ}}^2}\right)---\left(5\right)$

Indicate the likelihood function without target (only comprising noise), owing to the observed value of each pixel of image after pre-treatment is separate, then after pre-treatment image likelihood ratio function L (| z_k||x_k) be the product of all pixel observed value likelihood ratio functions, that is:

$L\left(\right|z_k\left|\right|x_k)=\underset{i=1}{\overset{N}{\mathrm{\Π}}}L\left(\right|z_k^i\left|\right|x_k)=\exp (-\frac{h{\left(x_k\right)}^{H}h\left(x_k\right)}{{2\mathrm{\σ}}^2})\underset{i=1}{\overset{N}{\mathrm{\Π}}}{I}_0\left(\frac{\left|h^i\left(x_k\right)z_k^i\right|}{{\mathrm{\σ}}^2}\right)---\left(6\right)$

Wherein, H represents the conjugate transpose of matrix, and N represents the number of the pixel of image after pre-treatment. The L that more than draws (| z_k||x_k) it is the amplitude likelihood ratio function that existing method draws

2) the present invention answers the setting of likelihood ratio function:

When targets are present, target signal is at the observed value z in k moment_kProbability density function p (z_k| target, φ) be:

$p\left(z_k\right|t\arg \mathrm{et},\mathrm{\φ})=\frac{1}{{\left|2\mathrm{\πR}\right|}^{1/2}}\exp \{-\frac{1}{2}{(z_k-s_{k}h\left(x_k\right))}^H{R}^{-1}(z_k-s_{k}h\left(x_k\right))\}---\left(7\right)$

In formula (7), H represents the conjugate transpose of matrix, and R is the covariance matrix of the white complex gaussian noise of the zero-mean of setting, and φ represents the phase place of target signal, | | represent determinant of a matrix, s_kFor the target signal vector in k moment, x_kRepresent the state of target in the k moment,

When target does not exist, target signal is at the observed value z in k moment_kProbability density function p (z_k| notarget) be:

$p\left(z_k\right|\mathrm{not}\arg \mathrm{et})=\frac{1}{{\left|2\mathrm{\πR}\right|}^{1/2}}\exp \{-\frac{1}{2}{z}_k^H{R}^{-1}{z}_k\}---\left(8\right)$

Thus likelihood ratio function L (z_k|x_k, φ) and it is p (z_k| target, φ) and p (z_k| notarget) ratio, then:

$\begin{array}{c}L\left(z_k\right|x_k,\mathrm{\φ})=\exp \{-\frac{1}{2}{(z_k-s_{k}h\left(x_k\right))}^H{R}^{-1}(z_k-s_{k}h\left(x_k\right))+\frac{1}{2}{z}_k^H{R}^{-1}{z}_k\}\\ =\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}\×\exp \{\frac{1}{2}{s}_k{z}_k^H{R}^{-1}h\left(x_k\right)+\frac{1}{2}{s}_k^{*}h{\left(x_k\right)}^H{R}^{-1}{z}_k\}\end{array}---\left(9\right)$

In (9) formula, subscript * represents the conjugation of matrix, then,

Make ξ=Eexp{j θ }=h (x)^HR^-1z(10)

By (10) and substitute into (9) formula:

$\begin{array}{c}L\left(z_k\right|x_k,\mathrm{\φ})=\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}\exp \{\frac{1}{2}{s}_k{\mathrm{\ξ}}^{*}+\frac{1}{2}{s}_k^{*}\mathrm{\ξ}\}\\ =\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}\×\exp \{\frac{1}{2}(\cos \mathrm{\φ}+j\sin \mathrm{\φ}){\mathrm{\ξ}}^{*}+\frac{1}{2}(\cos \mathrm{\φ}-j\sin \mathrm{\φ})\mathrm{\ξ}\}\\ =\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}\exp \{E\cos (\mathrm{\φ}-\mathrm{\θ})\end{array}---\left(11\right)$

(11) formula is sought marginal distribution, has:

$\begin{array}{c}L\left(z_k\right|x_k)={\∫}_0^{2\mathrm{\π}}L\left(z_k\right|x_k,\mathrm{\φ})p\left(\mathrm{\φ}\right)\mathrm{d\φ}\\ =\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}{\∫}_0^{2\mathrm{\π}}\frac{1}{2\mathrm{\π}}\×\exp \left\{\mathrm{Eos}(\mathrm{\φ}-\mathrm{\θ})\right\}\mathrm{d\φ}\end{array}---\left(12\right)$

Wherein, L (z_k|x_k) represent the k moment multiple likelihood ratio function, will (10) formula substitute into (12) formula, must:

$L\left(z_k\right|x_k)=\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}{I}_0\left(\right|h{\left(x_k\right)}^H{R}^{-1}{z}_k\left|\right)---\left(13\right)$

Wherein, I₀() is zeroth order Bessel's function; The multiple Gaussian noise of the zero-mean set when each moment mutual uncorrelated time, L (z_k|x_k) it is reduced to:

$L\left(z_k\right|x_k)=\exp (-\frac{h^H\left(x_k\right)h\left(x_k\right)}{{2\mathrm{\σ}}^2})I_0\left(\frac{\left|h^H\left(x_k\right)z_k\right|}{{\mathrm{\σ}}^2}\right)---\left(14\right)$

Wherein, H represents the conjugate transpose of matrix, σ²Represent the variance of the multiple Gaussian noise of the zero-mean that each moment sets.

The process of initialization process is below described:

Represent the initial moment with 1 moment, set backtracking function S (1) in 1 moment and the value function I (x in 1 moment₁|z₁):

I(x₁|z₁)=L (z₁|x₁)

S(x₁)=0

Wherein, I () representative value function, S () represents backtracking function.

S22: recurrence cumulative process: when k gets 2 to K, utilize following formula obtain the k moment accumulate after value function I (x_k|Z_1:K):

$I\left(x_k\right|X_{1:K})=\underset{x_{k-1}\∈\mathrm{\τ}\left(x_k\right)}{\max }[I\left(x_{k-1}\right|Z_{1:K})+\mathrm{Tr}\left(x_k\right|x_{k-1})]+L\left(z_k\right|x_k)$

Wherein, Z₁:_K={ z₁,z₂,…,z_K, τ (x_k) represent the state range that target state can shift in the k-1 moment; Tr (x_k|x_k-1) represent the penalty (any one penalty) that target state shifts; When k gets 2, I (x_k-1|Z_1:K)=I (x₁|Z₁)。

As k=K, draw I (x_K|Z_1:K), make I (x_K)=I (x_K|Z_1:K), according to aforementioned explanation, now I (x_K) it is only x_KFunction, τ (x_k) determine according to the motion characteristics of target.

S23: the x finding out satisfied setting condition_k, described setting condition is I (x_K)>V_DT, V_DTFor setting thresholding, such as, V_DTArrange according to given false-alarm rate.

By the x of satisfied setting condition_kNumber represent the x for Q, Q satisfied setting condition_kIt is expressed as:ExtremelyUtilizeExtremelyComposition state sequence

S24: respectively forExtremelyCarry out flight path backtracking, draw Q corresponding targetpath; The process drawing q targetpath is:

When k gets K-1 to 1, it may also be useful to following formula draws

Wherein, q gets 1 to Q, and S () represents backtracking function; Then utilizeExtremelyForm the q targetpath.

Upon step s 2, the target false track that causes of diffusion is also removed successively and false track that noise causes. It is described respectively below:

Remove target and spread the false track that causes: the false flight path that target diffusion causes has an obvious feature to be exactly these false flight paths with the flight path of living target and there is the track partially overlapped, remove, based on this principle, the false track that target diffusion causes. Specifically, after drawing Q targetpath, willExtremelyIt is substituting to respectively in value function I (xK), obtainsExtremelyWillExtremelyTargetpath corresponding to middle maximum value is designated as maximum value function target flight path, if there is M identical target state at remaining either objective flight path (non-maximum value function target flight path) and maximum value function target flight path, then the targetpath of correspondence is removed, obtain the targetpath after first pseudo-flight path removal. M be setting be greater than 1 natural number, the value of M depend on the time span of accumulation (value of K) and set thresholding V_DTSize.

Remove the false track that noise causes: after the targetpath after obtaining first pseudo-flight path and removing, in the targetpath after first pseudo-flight path is removed, for every bar targetpath, the target direction of motion of correspondence is carried out statistics with histogram; Then according to statistics with histogram result, the false track owing to noise causes is removed. For living target, its kinestate is unknown, but motion is regular, instead of random random motion, this target direction of motion shown on target track between contiguous frames will trend towards the total direction of motion determined by starting point and terminating point. For the false track caused by noise, owing to kinestate is random mixed and disorderly, performance in the movement direction adjacent interframe trend towards being uniformly distributed of-2 π～2 π. Therefore by the direction of motion of target is carried out statistics with histogram, it is possible to the false track that effective process causes due to noise.

Below by emulation experiment, the effect of the present invention is described further.

Emulation experiment content: image after the pre-treatment that utilization emulation produces, tests in MATLAB9.0 software, carries out detecting front tracking according to method provided by the invention, draw the flight path of target; Result as shown in Figure 2, with reference to Fig. 2 a, for utilizing the schematic diagram of targetpath that the present invention draws and true flight path.In Fig. 2 a, transverse axis representative frame number (i.e. k value), the longitudinal axis represents Range resolution unit number. With reference to Fig. 2 b, for utilizing the schematic diagram of targetpath and the true flight path drawn during amplitude likelihood ratio structure dynamic programming accumulating value function. In Fig. 2 b, transverse axis representative frame number (i.e. k value), the longitudinal axis represents Range resolution unit number. With reference to Fig. 2 c, it is the present invention and existing method (utilizing amplitude likelihood ratio structure dynamic programming accumulating value function, in figure 2 c referred to as amplitude likelihood ratio) the detection probability curve under different signal to noise ratio when false-alarm rate is identical. In Fig. 2 c, transverse axis represents signal to noise ratio, and unit is decibel; The longitudinal axis represents detection probability. With reference to Fig. 2 d, it is the present invention and existing method (utilizing amplitude likelihood ratio structure dynamic programming accumulating value function, in figure 2d referred to as amplitude likelihood ratio) the detecting and tracking probability curve under different signal to noise ratio when false-alarm rate is identical. In Fig. 2 d, transverse axis represents signal to noise ratio, and unit is decibel; The longitudinal axis represents tracking probability.

Emulation result analysis: table 1 represents the present invention and existing method (utilizing amplitude likelihood ratio structure dynamic programming accumulating value function) working time under identical simulating scenes.

Table 1

As seen from Table 1, the operation efficiency of the present invention is greatly improved than existing method. In addition from Fig. 2 c and Fig. 2 d it may be seen that the method that the present invention proposes is at detection perform or is all better than existing method on tracking performance, thus demonstrate the validity of the present invention.

Obviously, the present invention can be carried out various change and modification and not depart from the spirit and scope of the present invention by the technician of this area. Like this, if these amendments of the present invention and modification belong within the scope of the claims in the present invention and equivalent technologies thereof, then the present invention also is intended to comprise these change and modification.

Claims (5)

1. follow the tracks of method based on before the multiple Likelihood ration test of dynamic programming, it is characterised in that, comprise the following steps:

S1: utilize radar to receive radar return data, radar return data are carried out pre-treatment, obtain image after pre-treatment;

In step sl, after pre-treatment, the model representation of image is:

z_k=Aexp{j φ } h (x_k)+n_k

Wherein, n_kFor the white complex gaussian noise of the zero-mean that the k moment sets, A represents the amplitude of target signal, and φ represents the phase place of target signal, and φ obeys being uniformly distributed on [0,2 π], and h () represents point spread function, z_kFor target signal is at the observed value in k moment, x_kFor target is in the state in k moment; K gets 1 to K, K be setting be greater than 1 natural number;

S2: according to image after pre-treatment, adopts dynamic programming method to carry out detecting front tracking, draws the flight path of target; When following the tracks of before adopting dynamic programming method to detect, according to image after pre-treatment, it is to construct multiple likelihood ratio function, according to multiple likelihood ratio function, it is to construct the value function of dynamic programming method;

In step s 2, adopt dynamic programming method to carry out detecting front tracking to comprise the following steps;

S21: the multiple likelihood ratio function L (z that the k moment is set_k|x_k);

$L\left(z_k\right|x_k)=\exp \{-\frac{1}{2}h{\left(x_k\right)}^H{R}^{-1}h\left(x_k\right)\}{I}_0\left(\right|h{\left(x_k\right)}^H{R}^{-1}{z}_k\left|\right)$

Wherein, H represents the conjugate transpose of matrix, and R represents the white complex gaussian noise n of the zero-mean of setting_kCovariance matrix, I₀() represents zeroth order Bessel's function;

Represent the initial moment with 1 moment, set the value function I (x in 1 moment₁|z₁):

I(x₁|z₁)=L (z₁|x₁)

Wherein, I () representative value function;

S22: when k gets 2 to K, utilize following formula obtain the k moment accumulate after value function I (x_k|Z_1:K):

$I\left(x_k\right|Z_{1:K})=\underset{x_{k-1}\∈\mathrm{\τ}\left(x_k\right)}{max}\[I\left(x_{k-1}\right|Z_{1:K})+Tr\left(x_k\right|x_{k-1})\]+L\left(z_k\right|x_k)$

Wherein, Z_1:K={ z₁, z₂..., z_K, τ (x_k) represent that the k-1 moment can transfer to state x_kAll state x_k-1;Tr (x_k|x_k-1) represent the penalty that target state shifts; When k gets 2, I (x_k-1|Z_1:K)=I (x₁|Z₁);

As k=K, draw I (x_K|Z_1:K), make I (x_K)=I (x_K|Z_1:K);

S23: the x finding out satisfied setting condition_k, described setting condition is I (x_K) > V_DT, V_DTFor setting thresholding; Meet the x of setting condition_kNumber represent the x for Q, Q satisfied setting condition_kIt is expressed as:ExtremelyUtilizeExtremelyComposition state sequence

S24: respectively forExtremelyCarry out flight path backtracking, draw Q corresponding targetpath; The process drawing q targetpath is:

When k gets K-1 to 1, it may also be useful to following formula draws

Wherein, q gets 1 to Q, and S () represents backtracking function; Then utilizeExtremelyForm the q targetpath,Represent that in the q targetpath, target is in the state in k moment.

2. follow the tracks of method based on before the multiple Likelihood ration test of dynamic programming as claimed in claim 1, it is characterized in that, in step sl, radar return data being carried out pretreated process is: radar return data are carried out clutter recognition, squelch and process of pulse-compression.

3. follow the tracks of method based on before the multiple Likelihood ration test of dynamic programming as claimed in claim 1, it is characterised in that, in the step s 21, the multiple Gaussian noise of the zero-mean set when each moment mutual uncorrelated time, L (z_k|x_k) it is reduced to:

$L\left(z_k\right|x_k)=\exp (-\frac{h^H\left(x_k\right)h\left(x_k\right)}{2{\mathrm{\σ}}^2})I_0\left(\frac{\left|h^H\left(x_k\right)z_k\right|}{{\mathrm{\σ}}^2}\right)$

Wherein, H represents the conjugate transpose of matrix, σ²Represent the variance of the multiple Gaussian noise of the zero-mean that each moment sets, I₀() represents zeroth order Bessel's function.

4. follow the tracks of method based on before the multiple Likelihood ration test of dynamic programming as claimed in claim 1, it is characterised in that, upon step s 2, after drawing Q targetpath, willExtremelyIt is substituting to value function I (x respectively_K) in, obtainExtremelyWillExtremelyTargetpath corresponding to middle maximum value is designated as maximum value function target flight path, if there is M identical target state at remaining either objective flight path and maximum value function target flight path, then the targetpath of correspondence is removed, obtain the targetpath after first pseudo-flight path removal; M be setting be greater than 1 natural number.

5. follow the tracks of method based on before the multiple Likelihood ration test of dynamic programming as claimed in claim 4, it is characterized in that, after targetpath after obtaining first pseudo-flight path and removing, in targetpath after first pseudo-flight path is removed, for every bar targetpath, the target direction of motion of correspondence is carried out statistics with histogram; Then according to statistics with histogram result, the false track owing to noise causes is removed.